Cantrell, Martin and Ellis present a compelling consideration of the prospects for automated curation of wild places, i.e. the development of deep-learning systems that can conserve and maintain wild places with minimal human intervention. The most intriguing part of this paper is the “wildness creator,” which is:
… a conceptual design for an autonomous landscape infrastructure system that creates and sustains wildness by enhancing nonhuman influences while countering all forms of human influence. It is a deep learning computing system that controls a physical infrastructure that can sense and manipulate the environment and interact with organisms. (p. 163)
To achieve these goals, “the system would learn the most effective strategies.” (ibid.) In other words, the wildness creator has an interface that enables it to control and manipulate the physical world, while its behavior is governed by a machine learning algorithm. This conceptual design is far beyond any existing autonomous systems, but even the prospect of such a system raises difficult questions about how we define conservation goals, whether there ought to be any constraints on the behavior of this system, and even the prospect of the co-evolution of the wildness creator and the organisms it manages.
Consider a different autonomous system that we all have some familiarity with: the driverless car. The goal of this autonomous system is well defined (get from one place to another), and so are the constraints on its behavior (stay on the road, follow the rules of the road, don’t run into other cars, etc.).
In the case of the wildness creator, neither the goal nor the constraints on its behavior are well specified. What would the constraints on the behavior of the wildness creator be? As an example, imagine that the wildness creator is managing a community where a single species is becoming disproportionately abundant. Would it be permissible for it to cull individuals of that species in pursuit of a more stable community? Could it manipulate the fecundity or mortality of certain individuals or species? These are obvious and likely scenarios, and the prospect of an ecosystem governed by an autonomous system raises complicated ethical questions about what actions the wildness creator ought to be allowed to take.
More fundamentally, what goal or end-state would the wildness creator pursue? Conservation biology offers compelling arguments for many different conservation goals: conservation of rare species, maintenance of the abundance of common species, maximization of the delivery of ecosystem services, maintenance of a particular assemblage of species, etc. As Galaz et al. (see Further Reading below) point out in a response to Cantrell et al., choosing among these goals is a fundamentally value-laden process. Thus, is it even possible for humans or the wildness creator itself to formulate goals that are value-neutral? Would even be possible to allow the wildness creator to define its own goals?
Finally, the prospect of an autonomous system manipulating an ecosystem raises the possibility of co-evolution of the wildness creator and the species it manages. In ecology, co-evolution refers to the case where closely associated (i.e., interacting) species reciprocally affect each other’s evolution. Could the wildness creator and the organisms it manages co-evolve? The wildness creator is by design a system that evolves, and there are also many examples of rapid evolution of species in response to human actions (e.g., change in length and age of maturity for fish in response to fishing pressure). Presumably, then, species could evolve rapidly in response to actions of the wilderness creator. Given that we have two highly interactive sets of agents (organisms and the wildness creator), both with the capacity to adapt and evolve in response to the actions of the other, co-evolution would almost certainly occur.
Co-evolution of the wildness creator and the organisms it manages would offer a unique avenue for niche construction: species could evolve the ability to manipulate the wildness creator to modify their environment for them (a sort of indirect niche construction). Presumably, fostering this type of rapid evolution of species contradicts the ethos of minimum interference that inspired the wildness creator in the first place. Would the wildness creator be designed in a way to prohibit co-evolution? Would that be possible?
Given the rapid development of autonomous systems in many different fields, this paper by Cantrell et al. is timely and raises intriguing questions about the future role of autonomous systems in conservation.
Galaz, V. and Mouazen, A.M. 2017. “‘New Wilderness’ Requires Algorithmic Transparency: A Response to Cantrell et al. Trends in Ecology and Evolution, in press (DOI: http://dx.doi.org/10.1016/j.tree.2017.06.013). A letter in TREE on the Cantrell paper.
Inhabiting the Anthropocene 
Thanks for this Tom–what a fascinating article! Two quick comments (more like headlines for longer comments!)
First, is what the “wildness creator” produces wilderness? I guess Cantrell et al. talk about the distinction a bit, but there’s a lot more to consider here . . . in particular, what’s the relationship between the present absence of (detectable?) human influence, and the history of what actual processes led to the current condition of a given place. If there’s a meaningful difference, why does it matter? Wow, there’s a lot to think about here . . .
Second, and actually related, as a Rousseau scholar I couldn’t help but think about the passage from Rousseau’s novel Julie about a garden (“Elysium”) she creates on the estate where she lives. It was a piece of wasteland, but by careful rerouting of a stream and judicious planting of native species she creates an apparent wilderness. The passage is incredibly rich and (I think) important–and it’s amazing how the description of the wildness creator echoes ideas from it. So now I’m thinking about a post on Elysium 2.0!